• Graduate program
    • Why Tinbergen Institute?
    • Program Structure
    • Courses
    • Course Registration
    • Facilities
    • Admissions
    • Recent PhD Placements
  • Research
  • News
  • Events
    • Events Calendar
    • Tinbergen Institute Lectures
    • Annual Tinbergen Institute Conference
    • Summer School
      • Inequalities in Health and Healthcare
      • Research on Productivity, Trade, and Growth
      • Behavioral Macro and Complexity
      • Business Data Science Summer School Program
    • Events Archive
  • Summer School
    • Inequalities in Health and Healthcare
    • Research on Productivity, Trade, and Growth
    • Behavioral Macro and Complexity
  • Alumni
  • Times
Home | Events Archive | Disentangling Exploration from Exploitation
Seminar

Disentangling Exploration from Exploitation


  • Series
  • Speaker(s)
    Alessandro Lizzeri (Princeton University, United States)
  • Field
    Organizations and Markets
  • Location
    Online
  • Date and time

    May 17, 2021
    16:00 - 17:00

Abstract:
A large literature in economics and statistics has studied a class of experimentation problems, often referred to as bandit problems. In this literature, at each date the DM’s receives payoffs from the project she chooses (or exploits) and learns (experiments) by observing these payoffs. Thus, exploration and experimentation are intimately connected. Specifically, the DM gains information only about the project she receives payoffs from. This gives rise to the well-known trade-off between exploration and exploitation that has two fundamental consequences: First, the DM may wish to invest in an ex-ante inferior project because this allows valuable learning about potentially superior rewards; Second, learning may be incomplete even in the long run because experimentation potentially requires receiving ex ante inferior payoffs for a long time. However, in many applications, exploration and exploitation are not as closely intertwined as these classical experimentation models suggest. For instance, individuals may monitor various stocks while invested in others, politicians may explore new policies while implementing old ones, employees may pursue on-the-job search. This paper investigates the consequences of a key modification to the canonical
exploration-exploitation environment. DMs can invest in a project while observing the returns of another. Put differently, we study the consequences of disentangling exploration from exploitation. We study the problem in a variation of the Poisson bandits model first introduced by Keller, Rady, and Cripps 2005. We fully characterize the optimal policy in our setting and show that it displays significant differences from the classical model. Joint work with Eran Shmaya and Leeat Yariv.

• If you want to attend this online seminar, you need to register here. You will then receive the details of the zoom session by email.

• Your microphone will be on mute upon joining the meeting, please leave it like that and unmute it only if you want to ask a question.

• Asking questions: please just go ahead and ask questions in the “usual way” (ie, don’t use the chat unless you want to notify the host of any problem related to seminar).

• Please use the registration form also to register for a Zoom bilateral on the day of the seminar. Deadline for requesting a bilateral is Thursday May 13 at 09:00.